Texture Atlas for more optimized instancing

D3D11Engine/BaseGraphicsEngine.h

@@ -230,4 +230,6 @@ class BaseGraphicsEngine {
     virtual XRESULT UpdateRenderStates() { return XR_SUCCESS; };
 
     virtual std::unique_ptr<GraphicsEventRecord> RecordGraphicsEvent( LPCWSTR region ) { return std::make_unique<GraphicsEventRecord>(); }
+
+    virtual void OnWorldLoaded() {};
 };

D3D11Engine/ConstantBufferStructs.h

@@ -16,17 +16,104 @@ struct VobInstanceInfo {
     DWORD GP_Slot;
 };
 
-/** Remap-index for the static vobs */
-struct VobInstanceRemapInfo {
-    bool operator < ( const VobInstanceRemapInfo& b ) const {
-        return InstanceRemapIndex < b.InstanceRemapIndex;
-    }
-
-    bool operator == ( const VobInstanceRemapInfo& o ) const {
-        return InstanceRemapIndex == o.InstanceRemapIndex;
-    }
-
-    DWORD InstanceRemapIndex;
+struct VobInstanceInfoAtlas {
+    XMFLOAT4X4 world;
+    XMFLOAT4X4 prevWorld;  // Previous frame's world matrix for motion vectors
+    DWORD color;
+    float windStrenth;
+    float canBeAffectedByPlayer;
+    // Texture Atlas information, directly stored in the instance data for easy access in shader without needing an extra StructuredBuffer
+    int slice;
+    float uStart;
+    float vStart;
+    float uEnd;
+    float vEnd;
+    UINT globalSourceIndex;  // global source index into feedback texture
+    float minHeight;         // BBox.Min.y for per-vob wind calculations
+    float maxHeight;         // BBox.Max.y for per-vob wind calculations
+};
+
+// Descriptor returned for use with shader
+// Points to a specific slice in the Texture2DArray atlas, along with UV coordinates for sampling that slice
+// this is pointed to from VobInstanceInfo GP_Slot into a StructuredBuffer, which is then indexed in the shader to get the correct slice/UVs for each instance
+struct TextureDescriptor {
+    int slice;
+    float uStart;
+    float vStart;
+    float uEnd;
+    float vEnd;
+};
+
+// CPU-side lookup: maps a zCTexture* to its atlas placement
+struct TextureAtlasLookup {
+    DXGI_FORMAT atlasFormat;
+    TextureDescriptor descriptor;
+};
+
+// Per-vob data uploaded once at world load, read by GPU cull compute shader
+struct VobGPUData {
+    XMFLOAT3 aabbCenter;
+    float pad0;
+    XMFLOAT3 aabbExtent;
+    float pad1;
+    XMFLOAT4X4 world;
+    XMFLOAT4X4 prevWorld;
+    DWORD color;
+    float aniModeStrength;
+    float canBeAffectedByPlayer;
+    UINT submeshStart;       // index into SubmeshGPUData[]
+    UINT submeshCount;       // how many submeshes this vob maps to
+    float minHeight;         // BBox.Min.y for per-vob wind calculations
+    float maxHeight;         // BBox.Max.y for per-vob wind calculations
+    UINT pad2;
+};
+
+// Per-submesh lookup, shared across all vobs with the same visual
+struct SubmeshGPUData {
+    int slice;
+    float uStart, vStart, uEnd, vEnd;
+    UINT argIndex;              // index into merged indirect args
+    UINT instanceBaseOffset;    // fixed write offset in instance buffer
+    UINT globalSourceIndex;     // global source index into feedback texture
+};
+
+// Per-submesh data for the world mesh atlas indirect draw path.
+// Read by VS_ExWorldAtlas via StructuredBuffer<WorldMeshSubmeshGPUData>.
+struct WorldMeshSubmeshGPUData {
+    // Diffuse atlas
+    int   diffuseSlice;
+    float dUStart, dVStart, dUEnd, dVEnd;
+    // Normal atlas
+    int   normalSlice;
+    float nUStart, nVStart, nUEnd, nVEnd;
+    // FX atlas
+    int   fxSlice;
+    float fUStart, fVStart, fUEnd, fVEnd;
+    // Flags: 1 = HAS_NORMAL, 2 = HAS_FX, 4 = ALPHA_TEST
+    UINT  flags;
+};
+
+// Constant buffer for the GPU cull compute shader
+struct CullConstants {
+    XMFLOAT4 frustumPlanes[6];
+    XMFLOAT3 cameraPosition;
+    float drawDistance;
+    float globalWindStrength;
+    UINT windAdvanced;
+    UINT numVobs;
+    UINT feedbackFrameNumber;   // >0 = write feedback in CS; 0 = disabled (e.g. shadow pass)
+    UINT enableHiZ;             // 1 = Hi-Z occlusion culling enabled
+    UINT hiZMipCount;
+    float hiZWidth;             // Hi-Z mip 0 dimensions (full depth buffer size)
+    float hiZHeight;
+    XMFLOAT4X4 viewProjection; // Current frame view-projection matrix for Hi-Z reprojection
+};
+
+struct HiZBuildConstants {
+    UINT outputWidth;
+    UINT outputHeight;
+    UINT inputMipLevel;
+    UINT isCopyPass;    // 1 = copy from depth buffer (mip 0), 0 = downsample from previous mip
 };
 
 #pragma pack (push, 1)	

D3D11Engine/D3D11AtlasTypes.h

@@ -0,0 +1,31 @@
+#pragma once
+#include "D3D11TextureAtlasManager.h"
+#include "D3D11IndirectBuffer.h"
+#include "ConstantBufferStructs.h"
+
+#include <d3d11.h>
+#include <vector>
+#include <memory>
+
+// Shared atlas constants
+constexpr size_t TEXTURE_ATLAS_MAX = DXGI_FORMAT_V408 + 1;
+struct MeshVisualInfo;
+
+// Tracks one unique submesh in the global geometry buffer
+struct StaticSubmeshEntry {
+    UINT indexCount;
+    UINT startIndexLocation;   // offset into global IB
+    int  baseVertexLocation;   // offset into global VB
+    TextureDescriptor atlasDesc;
+    MeshVisualInfo* visual;    // which visual owns this submesh
+};
+
+// Groups all submeshes that share one atlas (same DXGI_FORMAT)
+struct AtlasDrawGroup {
+    DXGI_FORMAT format = DXGI_FORMAT_UNKNOWN;
+    std::vector<StaticSubmeshEntry> submeshes;
+    std::vector<D3D11_DRAW_INDEXED_INSTANCED_INDIRECT_ARGS> indirectArgs;
+    std::unique_ptr<D3D11IndirectBuffer> indirectBuffer;
+    UINT mergedArgsOffset = 0;  // byte offset into merged indirect args buffer
+    UINT mergedArgsCount = 0;   // number of args in this group
+};

D3D11Engine/D3D11CascadedShadowMapBuffer.cpp

@@ -115,13 +115,13 @@ ID3D11ShaderResourceView* D3D11CascadedShadowMapBuffer::GetShaderResourceView()
     return m_srv.Get();
 }
 
-void D3D11CascadedShadowMapBuffer::BindToPixelShader( ID3D11DeviceContext1* context, UINT slot ) const {
+void D3D11CascadedShadowMapBuffer::BindToPixelShader( ID3D11DeviceContext* context, UINT slot ) const {
     if ( m_srv ) {
         context->PSSetShaderResources( slot, 1, m_srv.GetAddressOf() );
     }
 }
 
-void D3D11CascadedShadowMapBuffer::BindToVertexShader( ID3D11DeviceContext1* context, UINT slot ) const {
+void D3D11CascadedShadowMapBuffer::BindToVertexShader( ID3D11DeviceContext* context, UINT slot ) const {
     if ( m_srv ) {
         context->VSSetShaderResources( slot, 1, m_srv.GetAddressOf() );
     }

D3D11Engine/D3D11CascadedShadowMapBuffer.h

@@ -50,14 +50,14 @@ class D3D11CascadedShadowMapBuffer {
      * @param context Device context
      * @param slot Shader resource slot
      */
-    void BindToPixelShader( ID3D11DeviceContext1* context, UINT slot ) const;
+    void BindToPixelShader( ID3D11DeviceContext* context, UINT slot ) const;
 
     /**
      * Bind the texture array to a vertex shader slot.
      * @param context Device context
      * @param slot Shader resource slot
      */
-    void BindToVertexShader( ID3D11DeviceContext1* context, UINT slot ) const;
+    void BindToVertexShader( ID3D11DeviceContext* context, UINT slot ) const;
 
     /** Get the size of each cascade (width = height) */
     UINT GetSize() const { return m_size; }

D3D11Engine/D3D11CommandList.cpp

@@ -0,0 +1,11 @@
+#include "pch.h"
+#include "D3D11CommandList.h"
+#include "D3D11VertexBuffer.h"
+
+void D3D11CommandList::IASetVertexBuffer( D3D11VertexBuffer* vb, UINT stride, UINT offset ) {
+    m_Context->IASetVertexBuffers( 0, 1, vb->GetVertexBuffer().GetAddressOf(), &stride, &offset );
+}
+
+void D3D11CommandList::IASetIndexBuffer( D3D11VertexBuffer* ib, DXGI_FORMAT format, UINT offset ) {
+    m_Context->IASetIndexBuffer( ib->GetVertexBuffer().Get(), format, offset );
+}

D3D11Engine/D3D11CommandList.h

@@ -0,0 +1,140 @@
+#pragma once
+#include "pch.h"
+#include "D3D11PipelineStateObject.h"
+
+class D3D11VertexBuffer;
+
+/**
+ * Slim command-list wrapper around an ID3D11DeviceContext1 and the
+ * D3D11PipelineStateCache.
+ *
+ * Provides SetPipelineState() plus the commonly used Draw / IA / OM
+ * helpers so that call-sites read like a modern graphics API without
+ * touching the raw context or the engine's global render-state machine.
+ *
+ * The object is intentionally cheap to construct (two pointers) and
+ * does not own any resources.
+ */
+struct D3D11CommandList {
+
+    D3D11CommandList() = default;
+    D3D11CommandList( ID3D11DeviceContext* context, D3D11PipelineStateCache* cache )
+        : m_Context( context ), m_Cache( cache ) {}
+
+    // --- Pipeline state ------------------------------------------------------
+
+    void SetPipelineState( const D3D11PipelineStateObject& pso ) {
+        m_Cache->SetPipelineState( pso );
+    }
+
+    /** Force the cache to re-bind everything on next SetPipelineState. */
+    void InvalidatePipelineState() {
+        m_Cache->Invalidate();
+    }
+
+    // --- Input assembly ------------------------------------------------------
+
+    void IASetVertexBuffer( D3D11VertexBuffer* vb, UINT stride, UINT offset = 0 );
+
+    void IASetVertexBuffers( UINT startSlot,
+                             UINT numBuffers,
+                             ID3D11Buffer* const* buffers,
+                             const UINT* strides,
+                             const UINT* offsets ) {
+        m_Context->IASetVertexBuffers( startSlot, numBuffers, buffers, strides, offsets );
+    }
+
+    void IASetIndexBuffer( ID3D11Buffer* buffer, DXGI_FORMAT format, UINT offset = 0 ) {
+        m_Context->IASetIndexBuffer( buffer, format, offset );
+    }
+
+    void IASetIndexBuffer( D3D11VertexBuffer* ib, DXGI_FORMAT format, UINT offset = 0 );
+
+    // --- Draw calls ----------------------------------------------------------
+
+    void Draw( UINT vertexCount, UINT startVertexLocation = 0 ) {
+        m_Context->Draw( vertexCount, startVertexLocation );
+        m_DrawnTriangles += vertexCount / 3;
+    }
+
+    void DrawIndexed( UINT indexCount,
+                      UINT startIndexLocation = 0,
+                      INT  baseVertexLocation  = 0 ) {
+        m_Context->DrawIndexed( indexCount, startIndexLocation, baseVertexLocation );
+        m_DrawnTriangles += indexCount / 3;
+    }
+
+    void DrawInstanced( UINT vertexCountPerInstance,
+                        UINT instanceCount,
+                        UINT startVertexLocation    = 0,
+                        UINT startInstanceLocation  = 0 ) {
+        m_Context->DrawInstanced( vertexCountPerInstance, instanceCount,
+                                  startVertexLocation, startInstanceLocation );
+        m_DrawnTriangles += ( vertexCountPerInstance / 3 ) * instanceCount;
+    }
+
+    void DrawIndexedInstanced( UINT indexCountPerInstance,
+                               UINT instanceCount,
+                               UINT startIndexLocation   = 0,
+                               INT  baseVertexLocation    = 0,
+                               UINT startInstanceLocation = 0 ) {
+        m_Context->DrawIndexedInstanced( indexCountPerInstance, instanceCount,
+                                         startIndexLocation, baseVertexLocation,
+                                         startInstanceLocation );
+        m_DrawnTriangles += ( indexCountPerInstance / 3 ) * instanceCount;
+    }
+
+    void DrawIndexedInstancedIndirect( ID3D11Buffer* argsBuffer,
+                                       UINT alignedByteOffsetForArgs ) {
+        m_Context->DrawIndexedInstancedIndirect( argsBuffer, alignedByteOffsetForArgs );
+        // Triangle count unknown for indirect draws
+    }
+
+    // --- Render target / viewport helpers ------------------------------------
+
+    void OMSetRenderTargets( UINT numViews,
+                             ID3D11RenderTargetView* const* rtvs,
+                             ID3D11DepthStencilView* dsv ) {
+        m_Context->OMSetRenderTargets( numViews, rtvs, dsv );
+    }
+
+    void RSSetViewports( UINT numViewports, const D3D11_VIEWPORT* viewports ) {
+        m_Context->RSSetViewports( numViewports, viewports );
+    }
+
+    void RSGetViewports( UINT* numViewports, D3D11_VIEWPORT* viewports ) {
+        m_Context->RSGetViewports( numViewports, viewports );
+    }
+
+    void ClearDepthStencilView( ID3D11DepthStencilView* dsv,
+                                UINT clearFlags,
+                                float depth,
+                                UINT8 stencil ) {
+        m_Context->ClearDepthStencilView( dsv, clearFlags, depth, stencil );
+    }
+
+    void ClearRenderTargetView( ID3D11RenderTargetView* rtv, const float color[4] ) {
+        m_Context->ClearRenderTargetView( rtv, color );
+    }
+
+    // --- Stats ---------------------------------------------------------------
+
+    /** Return triangles drawn since last ResetStats() and reset counter. */
+    UINT FlushDrawnTriangles() {
+        UINT t = m_DrawnTriangles;
+        m_DrawnTriangles = 0;
+        return t;
+    }
+
+    UINT GetDrawnTriangles() const { return m_DrawnTriangles; }
+
+    // --- Raw access (escape hatch) -------------------------------------------
+
+    ID3D11DeviceContext*    GetContext() const { return m_Context; }
+    D3D11PipelineStateCache* GetPSOCache()   const { return m_Cache; }
+
+private:
+    ID3D11DeviceContext*    m_Context        = nullptr;
+    D3D11PipelineStateCache* m_Cache          = nullptr;
+    UINT                     m_DrawnTriangles = 0;
+};

D3D11Engine/D3D11Effect.cpp

@@ -219,15 +219,12 @@ XRESULT D3D11Effect::DrawRain() {
     // Set alphablending
 
     state.BlendState.SetAlphaBlending();
-    state.BlendState.SetDirty();
 
     // Disable depth-write
     state.DepthState.DepthWriteEnabled = false;
-    state.DepthState.SetDirty();
 
     // Disable culling
     state.RasterizerState.CullMode = GothicRasterizerStateInfo::CM_CULL_NONE;
-    state.RasterizerState.SetDirty();
 
     // Rendering instances only
     e->GetContext()->IASetPrimitiveTopology( D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP );
@@ -376,15 +373,12 @@ XRESULT D3D11Effect::DrawRain_CS() {
     // Set alphablending
 
     state.BlendState.SetAlphaBlending();
-    state.BlendState.SetDirty();
 
     // Disable depth-write
     state.DepthState.DepthWriteEnabled = false;
-    state.DepthState.SetDirty();
 
     // Disable culling
     state.RasterizerState.CullMode = GothicRasterizerStateInfo::CM_CULL_NONE;
-    state.RasterizerState.SetDirty();
 
     // Rendering instances only
     e->GetContext()->IASetPrimitiveTopology( D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP );